labor market policies, institutions and employment rates in the eu-27

Labour Market Policies and Outcomes in theEnlarged EU*jcms_2068 661..686

RANDOLPH L. BRUNOUniversità di Bologna

RICCARDO ROVELLIUniversità di Bologna

Abstract

We document and compare labour market institutions, policies and outcomes in theEU Member States, for the period between 1999 and 2006. Higher employment ratesare in general positively associated with measures of policy generosity, especiallywith the use of active labour market policies (ALMP), and negatively with institu-tions and policies which induce rigidity in the labour market. We also find evidencethat the relation between ALMP and employment levels is non-monotonic and that itis conditional on the informal institutions prevailing in different countries.

Introduction

The performance of labour markets has been at the centre of many analysesand debates on the European economies. In the 1980s, the concept of ‘euro-sclerosis’ was often used to refer to the combination of slow growth, lowemployment creation and rigid labour markets which then characterizedmany ‘old’ European countries. The functioning of labour markets has beena constant preoccupation of European policy-makers, and it is also at the heartof the Lisbon strategy (which the European Council adopted in March 2000)

* We thank two anonymous referees and, for helpful comments on a previous version, Giorgio Bellettini,Andrea Ichino, Hartmut Lehmann, Anzelika Zaiceva and other participants at seminars at the Università diBologna, Charles University (Prague, CZ) and the London School of Economics (UK). This article is asubstantially revised version of Rovelli and Bruno (2008).

JCMS 2010 Volume 48. Number 3. pp. 661–685

© 2010 The Author(s)Journal compilation © 2010 Blackwell Publishing Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148,USA

and of its reappraisal in 2005. One main objective of that strategy was to raisethe employment rate ‘as close as possible to 70 per cent by 2010’ in all the EUMember States – a goal that all the EU-15 Member States, but those who hadalready achieved it by 2000, were likely to miss even before the current crisis.

While well-functioning labour markets are a key ingredient to a successfuleconomic performance, labour market institutions also lie at the heart of thesocial models that so peculiarly characterize many EU Member States.1

Social policies clearly matter to the EU citizens, and citizens often out-spokenly support their own country’s labour market and social policies. In arecent Eurobarometer Survey on ‘European social reality’ (2007), 51 per centof European Union citizens declared their satisfaction with the quality of thesocial welfare system in their own country.2

However, shared values and common objectives do not necessarily implythat policy competences should be allocated at the EU level and require EUlegislative action. In fact, and in accord with the principle of subsidiarity,most competences in the field of social and labour market policies are leftwith individual Member States. Thus, institutional pressure to adopt uniformlabour markets and social policies within the EU is quite low, and theobserved cross-country differences are likely to persist for a long time, unlessMember States spontaneously choose to learn from the best practice of othersand thus to converge, possibly towards a single common model.

In this article we characterize the main cross-country differences in labourmarket policies across EU Member States, and relate various measures andpolicy indicators to indicators of labour market outcomes. The focus of ourwork is a comparative analysis of how employment rates in each country arerelated to indicators of labour market institutions and policies, and in particu-lar to expenditures on labour market programmes (LMP).

The article is organized as follows. In section I we review the literature onthe links between labour market institutions, policies and outcomes. Insection II we motivate our empirical analysis and discuss our results, subject-ing them to a number of robustness checks. In the final section we offer someconcluding remarks. A short Appendix describes the data and presents someadditional results.

1 For a concise characterization of the differences within EU labour market and social models, see Rovelli(2008).2 The specific question asked was whether the welfare system in their country ‘provides wide enoughcoverage’ (Eurobarometer, 2007, p. 76). Countries where respondents expressed the least support for theirwelfare systems were Bulgaria (2 per cent), Portugal (5 per cent), Latvia (6 per cent), Romania (7 per cent)and Greece (8 per cent). In general, the survey observes that ‘people’s propensity to feel that their country’ssocial welfare system could serve as a model for other countries is strongly related to whether they feel itprovides enough coverage’ (Eurobarometer, 2007, p. 77).

662 RANDOLPH L. BRUNO AND RICCARDO ROVELLI

© 2010 The Author(s)Journal compilation © 2010 Blackwell Publishing Ltd

I. Institutions, Policies and Outcomes: A Review

The diversity within European countries of labour markets institutions andpolicies on the one hand and of outcomes on the other has been documentedand analysed in several papers and with different approaches. While someanalyses also introduce cyclical factors and the role of adverse economicshocks in raising unemployment levels, most explanations focus on the rolesof institutions and policies, and then additionally on their interactions withadverse shocks.

In one of the most cited comparative papers on the subject, Nickell (1997)studies the evolution of unemployment rates in 20 OECD countries, of which15 were in Europe, in 1983–94. He finds that in general high unemploymentis associated with welfare systems that do not put pressure on the unemployedto search for and accept work offers, with no co-ordination in wage bargain-ing and also with high taxes on labour and with poor educational standards atthe bottom end of the market. Interestingly, strict employment protection, thedegree of unionization and the generosity of benefits per se do not seem toaffect unemployment. This result suggests that it is really the pressure toaccept job offers – which may or may not be associated with the supply ofunemployment benefits – that has an impact on the diversity of unemploy-ment rates.

Somewhat in the same vein, Blanchard and Wolfers (2000) study theevolution of unemployment in 15 European countries over the 1960–94period. They also take into account the role of adverse and persistent macro-economic shocks, caused by changes in oil prices, productivity and monetaryand fiscal policies. By themselves, these shocks are not sufficient to explaineither the persistence of high unemployment or the differences in cross-country unemployment rates. It is really the interactions of shocks and insti-tutions (such as replacement rates, the duration of unemployment benefits, thelevel of employment protection laws – EPL – the tax wedge, and also unioncontract coverage and density) that lead to a larger effect of shocks onunemployment. On the other hand, other institutions, such as active labourmarket policies (ALMP) and more co-ordination in wage bargaining, have theopposite effects – that is, they dampen the effects of shocks on labour marketoutcomes.

Bassanini and Duval (2006) reappraise the same set of hypotheses, usinga more recent and wider data set of 21 OECD countries over the period1982–2003. They report that ‘high and long-lasting unemployment benefits,high tax wedges and stringent anti-competitive product market regulation arefound to increase aggregate unemployment’, and also to decrease labourmarket participation, particularly for groups of individuals located at the

LABOUR MARKET POLICIES AND OUTCOMES IN THE ENLARGED EU 663


margins of the labour market. ‘By contrast, highly centralized and/orco-ordinated wage bargaining systems are estimated to reduce unemploy-ment.’ In general, ‘changes in policies and institutions appear to explainalmost two-thirds of non-cyclical unemployment changes over the past twodecades’. According to Bassanini and Duval, the effects of different institu-tions and policies is quite large: ‘On average [. . .] a 10 percentage pointreduction in the tax wedge, a 10 percentage point reduction of unemploymentbenefits and/or a decline in product market regulation by two standard devia-tions would be associated with a drop in the unemployment rate by about 2.8,1.2 and 0.7 percentage points, respectively’. In addition, they also documentrelevant interactions between policies and institutions, and observe that ‘theimpact of generous unemployment benefits on unemployment appears to bemitigated by high public spending on ALMPs, perhaps because high spendingon ALMPs is often accompanied with a strong emphasis on “activation” ’,and also that ‘structural reforms appear to have mutually reinforcing effects:the impact of a given policy reform is greater the more employment-friendlythe overall policy and institutional framework’.

Contributions from political scientists and sociologists often adopt a dif-ferent conceptual set-up and a different format for the analysis, but theiroverall conclusions are not radically different from the economists’. Forinstance Bradley and Stephens (2007) appear to share many conclusionsreached by Bassanini and Duval (2006); however they interpret their findingsin relation to a wider debate between the neoliberal and institutionalapproaches. In their study, Bradley and Stephens also focus on the effects ofpolicies and institutions on employment levels, in relation to 17 advancedindustrial countries from 1974 to 1999. They assume Denmark and the UnitedStates as two opposite benchmarks, representing ‘alternative paths to highemployment’. The contrast arises from the fact that ‘whereas the neoliberalpolicy prescriptions are best represented by the United States, Denmark bestrepresents the policy package suggested by our analysis: very low socialsecurity taxes, modest EPL, strong ALMP effort, high levels of wageco-ordination and neocorporatism, and high short-term unemploymentreplacement rates’ (Bradley and Stephens, 2007). They also acknowledge thatin reality, the two views share at least some common middle ground, forwhich they indeed find supportive evidence. More precisely they confirm the‘neoliberal view’ that high social security taxes, high level and long durationof unemployment benefits3 and strict EPL tend to reduce employment rates.However, on all of these points, the neoliberals and institutionalists share the

3 On unemployment benefits, Howell et al. (2007, p. 40) suggest instead that the positive correlation oftenobserved between unemployment rates and gross benefit replacement ratios should be interpreted ascausality running from the former to the latter.



same view. On the other hand several other neoliberal hypotheses found nosupport in their analysis: ‘Consistent with the institutionalist view and con-trary to the neoliberal view, we found that neocorporatism, short-term unem-ployment replacement rates, and ALMP were strongly related to employmentlevels, and total taxes were not related to employment levels’ (Bradley andStephens, 2007).

Hence they end up finding broad confirmation for the institutionalisthypothesis, with two minor exceptions: ‘the hypotheses derived from insti-tutionalist work that we did not find support for were ones in which theyshared a common prediction with neoliberal economics. We found that uniondensity had no effect when bargaining arrangements were controlled for andthat wage dispersion had no effect’ (Bradley and Stephens, 2007, p. 1504).

Following a different sociological approach, Hicks and Kenworthy (2003)reappraise the analysis of Esping-Andersen (1990), who had identified threedimensions of the welfare states in the developed countries. On the basis of aprincipal component analysis, Hicks and Kenworthy (2003) suggest that it isconvenient to summarize the varieties of welfare states along two dimensions,that of ‘progressive liberalism’ and of ‘traditional conservatism’. At the oneextreme of each dimension they place respectively Denmark, Norway andSweden (for progressive liberalism) and Austria, Belgium, France, Germanyand Italy (for traditional conservatism). Interestingly, the US and Australiatend to be placed towards the opposite extremes of both dimensions (p. 33).They then exploit both dimensions to explain employment performance4

between countries through an OLS regression, and find that, while progres-sive liberalism has no adverse impact on employment performance, tradi-tional conservatism does have such a negative impact, which the authorsattribute to the negative effect of ‘statist policies in more patriarchal, dirigiste,and Catholic welfare states’ (p. 53). However, their method of analysis doesnot allow to single out the employment effects due to each policy measure orinstitution. This need not be interpreted as a shortcoming, if one maintainsthat the effect of many policies on the employment level can be either positiveor negative, depending on the overall policy context, and that policy comple-mentarities are quite common.

Boeri (2002), following Ferrera (1998), suggests an altogether differentreappraisal of the Esping-Andersen ‘varieties’. According to Boeri, fourmodels of social policy were prevailing in the EU at the end of the 20thcentury: the Nordic, Continental, Anglo-Saxon and Mediterranean models.While in abstract one could think that all social models want to pursue to

4 They also analyse the impact of these dimensions on ‘income redistribution’ and ‘gender equality in thelabour market’.



some extent three main tasks (to reduce poverty and income inequalities, toinsure against labour market risks, and to increase the incentives and rewardsfrom labour market participation) he argues that the latter task is crucial fordeveloped countries in the context of contemporary, global challenges. Sapir(2006), following along similar lines of reasoning, evaluates the performanceof the four models according to the third criterion, that is how they stimulatelabour market participation. He observes that the performance of the fourmodels can be usefully compared ‘with a typology based on two criteria:efficiency and equity. A model will be considered efficient if it providessufficient incentives to work and, therefore, if it generates relatively highemployment rates. It will be deemed equitable if it keeps the risk of povertyrelatively low’. In his analysis, Nordic countries score well on both criteria,while continental countries provide equity (low poverty) at the price of lowefficiency (low employment levels). Quite the opposite for Anglo-Saxoncountries: they score well for efficiency but not for equity. Finally, Mediter-ranean countries score badly on both accounts.

Diverse as they may be, these analyses have a lot in common: they allsuggest that there is a set of policies which, if properly enacted, contributes toincrease the opportunities and rewards from labour market participation,while at the same time providing adequate or possibly generous insuranceagainst labour market risks. Such policies seem to have found a more suc-cessful implementation in those countries which fall within the boundaries ofSapir’s ‘Nordic’ model or of Hicks and Kenworthy’s ‘progressive liberalism’.And this conclusion is broadly coherent, at least prima facie, with the analy-ses of Nickell (1997) and of Blanchard and Wolfers (2000).

There is also a widespread agreement on the existence of a negativerelationship between the strictness of EPL and higher unemployment or loweremployment levels: ‘most of the individual country studies demonstratethat regulations promoting job security [. . .] on net reduce employment’(Heckman and Pagés, 2003).

Nevertheless, this leaves still ample room for more specific debates. Inparticular, as Dar and Tzannatos (1999) observe: ‘there are polar positions onthe effectiveness of active labour market programmes. On one hand, propo-nents of these programmes argue that active labour market programmes arenecessary and useful, short only of a panacea for reducing unemployment andprotecting workers. Opponents of the programmes tend to summarily dismissthese programmes as a waste of public money with high opportunity costs toother social programmes and labour market efficiency as a whole’.

And on the same point, Betcherman et al. (2004) observe that ‘someALMPs do have positive impacts, with favorable cost–benefit ratios. How-ever, in many cases, programmes have not improved the future employment



prospects of participants and, when they have, they have not always done soin a cost-effective manner [. . .]. Employment services are generally the mostcost-effective intervention [. . .] and, compared to other ALMPs, [. . .] areinexpensive. Training programmes for the unemployed can also have positiveimpacts on employment [. . .]. These programmes are most effective whenthey are workplace-based’.

In the empirical analysis that follows, we shall also mainly focus onmodelling employment rates, as it is done in most of the cited publications. Inparticular, we expect to find that employment rates are positively related tothe adoption of active LMP, although the size of the observed correlations andthe cost-effectiveness of such programmes may well vary according to thetype of programme, to the cyclical conditions prevailing at each timeand to the degree of ‘employment-friendliness’ provided by other relevantpolicies and institutions adopted in each country.

II. Employment Rates in the Enlarged EU: An Empirical Analysis

In this section we present our empirical analysis. It first compares labourmarket institutions, policies and outcomes in the EU-27. Secondly, we moti-vate and discuss several regression models for the employment rates, andthirdly we provide several robustness checks.

A First Look at Policies and Outcomes

In Tables 1 and 2 we report summary statistics for labour market expendi-tures, institutional rigidities and labour market outcomes, respectively. Datacover all the EU Member States for which they are available, between 1999and 2006. Using data for the previous years would have limited the study toEU-15 Member States, thus largely replicating the studies reviewed in sectionI. On the contrary, one of our main points of interest is to determine whetherand to what extent the conclusions of these studies still apply to the enlargedEU. In the tables, we only show figures for the beginning and end of theperiod.

The first six columns of Table 1 allow a cross-country comparison ofexpenditures on LMP as a ratio of GDP. We distinguish between expenditureson active LMP in columns 1 and 2 (programmes 2 to 7, according to theEurostat classification),5 passive LMP (8–9) in columns 3 and 4 and totalLMP (1–9) in columns 5 and 6. Denmark and the other Nordic countries(Sweden and Finland, and also the Netherlands) clearly stand out in terms ofpolicy generosity. At the other extreme, a few countries spend less than half

5 See the Appendix for a description of the dataset used.



Tabl

e1:

Lab

our

Mar

ket

Exp

endi

ture

san

dIn

stitu

tiona

lR

igid

ities

,199

9–20

06

idL

abou

rM

arke

tE

xpen

ditu

res

2–7

Lab

our

Mar

ket

Exp

endi

ture

s8–

9

Lab

our

Mar

ket

Exp

endi

ture

s1–

9

Rig

idit

yof

Em

ploy

men

tIn

dex

Tax

Wed

geU

nion

Den

sity

Rep

lace

men

tra

te

1999

2006

1999

2006

1999

2006

2003

2006

1999

2006

1999

2006

1999

2006

EU

(15

coun

trie

s)eu

150.

340.

521.

371.

251.

711.

9941

4238

4028

3657

57

Aus

tria

at0.

400.

551.

311.

391.

852.

1133

3741

4137

3555

55B

elgi

umbe

0.25

0.88

1.82

1.81

2.28

2.89

2720

4443

5555

6363

Bul

gari

abg

–0.

39–

0.18

–0.

6346

4737

31–

––

–C

ypru

scy

–0.

07–

0.66

–0.

76–

–23

24–

––

–C

zech

Rep

ublic

cz0.

110.

120.

280.

230.

460.

4820

2841

4133

2250

50G

erm

any

de1.

070.

622.

122.

103.

192.

9944

4440

4026

2361

61D

enm

ark

dk1.

831.

522.

572.

664.

404.

3416

1740

3774

7064

61E

ston

iaee

–0.

04–

0.07

–0.

13–

5840

3420

1150

50Sp

ain

es0.

630.

631.

451.

442.

082.

1762

6328

3216

1672

69Fi

nlan

dfi

0.90

0.73

2.33

1.69

3.34

2.55

4448

4342

7674

6160

Fran

cefr

1.05

0.68

1.52

1.40

2.74

2.32

5556

4342

88

7173

Gre

ece

gr0.

240.

060.

400.

40.

640.

4758

5837

3826

2545

48H

unga

ryhu

–0.

19–

0.36

–0.

6433

3443

3933

1747

43Ir

elan

die

0.87

0.46

1.11

0.87

1.98

1.57

2733

2925

4135

2930

Ital

yit

0.53

0.44

0.67

0.80

1.20

1.27

5354

4443

3634

5254

Lith

uani

alt

–0.

18–

0.13

–0.

4047

4839

3416

1425

25L

uxem

bour

glu

0.08

0.39

0.50

0.59

0.58

1.04

4444

3030

1313

6161

Lat

via

lv–

0.17

–0.

30–

0.54

6259

3734

1916

5050

Mal

tam

t–

0.07

–0.

41–

0.56

––

1922

––

––

Net

herl

ands

nl0.

920.

752.

011.

472.

932.

6942

4234

3425

2271

71Po

land

pl–

0.37

–0.

71–

1.17

3733

3634

2416

4752

Port

ugal

pt0.

310.

450.

811.

271.

231.

8556

5127

2923

2378

78R

oman

iaro

–0.

10–

0.28

–0.

4265

5148

42–

––

–Sw

eden

se1.

951.

131.

640.

963.

592.

2842

4349

4581

7878

77Sl

oven

iasi

–0.

18–

0.39

–0.

6757

5738

3841

4463

63Sl

ovak

iask

–0.

13–

0.34

–0.

6438

3937

3045

3064

64U

nite

dK

ingd

omuk

0.09

0.04

0.36

0.19

0.45

0.60

1314

2526

3029

4545

Not

es:

See

Tabl

eA

1fo

rso

urce

and

defin

ition

ofda

taon

Lab

our

Mar

ket

Exp

endi

ture

s(S

ourc

eE

uros

tat)

.All

Exp

endi

ture

sre

port

edas

%of

GD

P.E

xpen

ditu

res

2–7:

Act

ive

LM

P;E

xpen

ditu

res

8–9:

Pass

ive

LM

P;E

xpen

ditu

res

1–9:

Tota

lLM

P(i

nclu

des

also

exp.

inla

bour

mar

kets

ervi

ces)

.Cor

rela

tion

betw

een

Act

ive

and

Pass

ive

LM

P:93

%.

Exp

endi

ture

data

n.a.

in19

99–2

002

for

Bul

gari

a,C

ypru

s,E

ston

ia,H

unga

ry,L

ithua

nia,

Lat

via,

Mal

ta,P

olan

d,R

oman

ia,S

love

nia

and

Slov

akia

.For

Den

mar

kw

eus

e20

04da

tafo

r20

06,f

orG

reec

e20

05fo

r20

06,f

orC

zech

Rep

ublic

2002

for

1999

.R

epla

cem

entr

ate:

netv

alue

ofre

plac

emen

tben

efits

inth

ein

itial

phas

eof

unem

ploy

men

trel

ativ

eto

aver

age

wag

e,si

ngle

with

noch

ildre

n.W

eus

e20

01da

tafo

r199

9,20

04fo

r200

6(S

ourc

eO

EC

D).



Tabl

e2:

Lab

our

Mar

ket

Out

com

es,1

999–

2006

id19

9920

06D

elta

2006

–199

9E

mp.

Une

mp.

Par

t.E

mp.

Une

mp.

Par

t.E

mp.

Une

mp.

Par

t.

EU

(15

coun

trie

s)eu

1562

.58.

568

.366

.27.

771

.73.

7-0

.83.

4

Aus

tria

at68

.63.

971

.470

.24.

873

.71.

60.

92.

4B

elgi

umbe

59.3

8.5

64.8

61.0

8.3

66.5

1.7

-0.2

1.7

Bul

gari

abg

50.4

16.4

60.3

58.6

9.0

64.4

8.2

-7.4

4.1

Cyp

rus

cy65

.74.

969

.169

.64.

673

.03.

9-0

.33.

9C

zech

Rep

ublic

cz65

.68.

671

.865

.37.

270

.4-0

.3-1

.4-1

.4G

erm

any

de65

.28.

271

.067

.59.

874

.82.

31.

63.

8D

enm

ark

dk76

.05.

280

.277

.43.

980

.51.

4-1

.30.

4E

ston

iaee

61.5

11.3

69.3

68.1

5.9

72.4

6.6

-5.4

3.0

Spai

nes

53.8

12.5

61.5

64.8

8.5

70.8

11.0

-4.0

9.3

Finl

and

fi66

.410

.273

.969

.37.

775

.12.

9-2

.51.

1Fr

ance

fr60

.910

.468

.063

.89.

270

.32.

9-1

.22.

3G

reec

egr

55.9

12.0

63.5

61.0

8.9

67.0

5.1

-3.1

3.4

Hun

gary

hu55

.66.

959

.757

.37.

561

.91.

70.

62.

2Ir

elan

die

63.3

5.7

67.1

68.6

4.5

71.8

5.3

-1.2

4.7

Ital

yit

52.7

11.0

59.2

58.4

6.8

62.7

5.7

-4.2

3.4

Lith

uani

alt

61.7

13.7

71.5

63.6

5.6

67.4

1.9

-8.1

-4.1

Lux

embo

urg

lu61

.72.

463

.263

.64.

666

.71.

92.

23.

4L

atvi

alv

58.8

14.0

68.4

66.3

6.8

71.1

7.5

-7.2

2.8

Mal

tam

t54

.26.

758

.153

.67.

157

.7-0

.60.

4-0

.4N

ethe

rlan

dsnl

71.7

3.2

74.1

74.3

3.9

77.3

2.6

0.7

3.2

Pola

ndpl

57.6

13.4

66.5

54.5

13.9

63.3

-3.1

0.5

-3.2

Port

ugal

pt67

.44.

570

.667

.97.

873

.60.

53.

33.

1R

oman

iaro

63.2

7.1

68.0

58.8

7.3

63.4

-4.4

0.2

-4.6

Swed

ense

71.7

6.7

76.8

73.1

7.0

78.6

1.4

0.3

1.8

Slov

enia

si62

.27.

367

.166

.66.

070

.94.

4-1

.33.

8Sl

ovak

iask

58.1

16.4

69.5

59.4

13.4

68.6

1.3

-3.0

-0.9

Uni

ted

Kin

gdom

uk71

.05.

975

.571

.65.

475

.70.

6-0

.50.

2

Sour

ce:

Eur

osta

t.E

mpl

oym

ent

Rat

e=

em01

1(n

ewSI

code

tsie

m01

0),U

nem

ploy

men

tR

ate

=em

071

(new

SIco

dets

iem

110)

.Pa

rtic

ipat

ion

Rat

e=

Em

ploy

men

tra

te*

[1/(

1-

Une

mpl

oym

ent

rate

)].



the EU average of 2 per cent of GDP in total LMP, namely all the newMember States (those that entered the EU since 2004) with the exception ofPoland, and also Greece and the UK.6 Countries that engage in active pro-grammes also devote considerable resources to passive policies (unemploy-ment benefits and early retirement schemes): in 2006 the cross-countrycorrelation between the two types of expenditures was 93 per cent.

The rigidity of employment index7 also varies considerably across coun-tries (columns 7 and 8). Rigidity is lowest in Belgium, the Czech Republic,Denmark and the UK and highest both in some ‘old’ EU Member States(Spain, France, Greece, Italy and Portugal) and some of the new ones(Estonia, Latvia, Lithuania and Romania). The tax wedge is less variableacross countries (columns 9 and 10), whereas union density appears highespecially in Denmark, Finland and Sweden (columns 11 and 12).8 Short termreplacement rates9 range from a minimum of 25 per cent in Lithuania to 78per cent in Portugal and Sweden (columns 13 and 14). In general, all thesemeasures do not vary much in the period considered.

In Table 2 we report the rates of employment, unemployment and partici-pation in 1999 and 2006, as well as their period change. Employment rates inparticular have evolved quite differently across the EU-27 economies. This isalso shown in more detail in Figure 1. Employment rates have been high andin general constant in many countries, and especially Austria, Denmark, theNetherlands, Sweden and the UK have already reached the so-called ‘Lisbonobjective’ of 70 per cent. Spain and Bulgaria show the largest increases. Onthe other hand, employment levels are still low in Bulgaria, Hungary, Italy,Malta, Poland, Romania and Slovakia. Poland and Romania are especiallyworrying, since employment levels have been on average decreasing signifi-cantly during the period.

Following Sapir (2006), we also compare employment rates and povertylevels across countries and periods. This comparison can be taken as asynthetic representation of labour market efficiency versus equity. Figure 2documents this evolution between 1999 and 2006. For the EU-15, employ-ment rates have on average increased, while poverty rates have remained

6 It has been observed that the composition of social expenditures in the UK is different from that of othercountries, as more social expenditures take the form of non-labour related expenditures. However thisdifference does not affect our regression results.7 Available for 2003 and 2006, see the Appendix. The studies referred to in the literature review have usedthe EPL1 and EPL2 (employment protection legislation) indexes from the OECD. However this indicatoris missing for most new Member States. In any case, for those countries for which both measures areavailable, the correlation between EPL and the Rigidity index included in Table 1 is quite high.8 Checchi and Lucifora (2002) study the evolution and determinants of union density rates across 14 EUMember States.9 The literature distinguishes also between the effects of short and long rates. See for example Nickell et al.(2005).



Figu

re1:

Em

ploy

men

tR

ates

inth

eE

U-2

7,19

99–2

006

50607080 50607080 50607080 50607080 50607080

1998

2000

2002

2004

2006

1998

2000

2002

2004

2006

1998

2000

2002

2004

2006

1998

2000

2002

2004

2006

1998

2000

2002

2004

2006

1998

2000

2002

2004

2006

Aus

tria

Bel

gium

Bul

garia

Cyp

rus

Cze

ch R

epub

licD

enm

ark

Est

onia

Fin

land

Fra

nce

Ger

man

yG

reec

eH

unga

ry

Irel

and

Italy

Latv

iaLi

thua

nia

Luxe

mbo

urg

Mal

ta

Net

herla

nds

Pol

and

Por

tuga

lR

oman

iaS

lova

kia

Slo

veni

a

Spa

inS

wed

enU

nite

d K

ingd

om

Sour

ce:A

utho

rs’

calc

ulat

ions

.



constant. We may use the EU-15 averages for 2006 (vertical and horizontallines) to roughly divide the graph in four parts, where we expect to find(following the initial suggestion from Sapir, 2006) the Mediterranean andAnglo-Saxon countries in the top left and right, and the Continental andNordic countries in the bottom left and right, respectively. Figure 2 alsoallows us to track the changes occurred between 1999 and 2006. We observethree main ‘facts’:

• Several NMS initially are positioned in the ‘Mediterranean’ block, althoughthose with higher income per capita (Czech Republic and Slovenia, andalso Slovakia) belong with the ‘Continental’ block.

• Most countries move to the right (higher employment rates), while povertyrates remain largely constant,10 with a few exceptions. This movementmakes the ‘Mediterranean’ and the ‘Anglo-Saxon’ blocks closer to eachother, and the same goes for the ‘Continental’ and ‘Nordic’ blocks.

10 As the main purpose of this graph is to compare ‘outcomes’ in two different dimensions, we plot on thevertical axis the poverty rate after taxes and subsidies.

Figure 2: Employment Rate vs. Risk of Poverty

atbebg

cz

de

dk

eees

eu15

fi

fr

gr

hu

ieit

lt

lu

lvmt

nl

pl

pt

ro

se

si

uk

at

bebg

cy

cz

dedk

ee

es

eu15

fifr

gr

hu

ie

it lt

lu

lv

mt

nl

pl

ptro

sesisk

uk

51

01

52

02

5A

t ris

k of

pov

erty

ra

te a

fter

soci

al t

ran

sfe

rs (

%)

50 60 70 80Employment Rate

1999-2006 1999 2006

Poverty data not available for Cyprus and Slovakia in 1999.Poverty data for Bulgaria, Estonia, Latvia, Lithuania, Hungary, Malta, Poland and Slovenia in 1999 replaced with 2000.

Source: Authors’ calculations.



• Looking at individual countries, it is notable that Poland and Romania arethe only two countries that clearly move in both the ‘wrong’ directions, andfurther away from the EU-15 average.

As these data show, it is really the cross-country heterogeneity of employmentrates that characterizes the data, together with a tendency towards clusteringaround the higher employment levels. For this reason, and also since thepotentially explanatory variables are quite different across countries andmuch less across time (see Table 1), we shall focus on modelling employmentlevels in the regression analysis that follows.11

Modelling Employment Rates

The purpose of the regression analysis is to investigate the relationshipbetween the employment rate as a dependent variable and labour marketinstitutions and policies as independent variables. Nickell (1997) andBlanchard and Wolfers (2000), among others, have found that the follow-ing variables were relevant to explain labour market outcomes: rigidity ofemployment,12 replacement rates (measured as the ratio of benefits towages), duration of unemployment benefits, expenditure on active labourmarket policies, union density, coverage of union contracts, co-ordination ofwage bargaining and the tax wedge. We have collected these variables forthe largest number of EU countries and the longest time span available. Inthe end we restricted ourselves to the period 1999–2006 and to 23 coun-tries, for some of which we only have observations for the second half ofthe sample. We also decided not to use union coverage and co-ordination ofwage bargaining (see the Appendix for more details on data availability).Following the procedure outlined by Nickell (1997), we average theobservations in two sub-periods (1999–2002; 2003–06). We thus obtain a‘short’ two-period panel. We also include in the regressions the change ininflation rates between the beginning and the end of each sub-period, mea-sured at the country level, and a time dummy for the second period, tocontrol for the difference in business cycles and common macro shocks,respectively.

To obtain a first benchmark, we follow the literature on the varieties ofsocial models and labour market policies and regress employment rates on

11 On the other hand, since unemployment rates are considerably more variable across time than employ-ment rates, they are more naturally amenable to the kind of analysis developed by Blanchard and Wolfers(2000) and also by Nickell et al. (2005).12 See footnote 8. As a robustness check, we also use EPL2 as a regressor, for the subset of countries forwhich it is available. Results are reported in Table A3 and discussed in section 3.3, but they do not changeappreciably from those in Table 3.



four geographical dummies, representing the four geographical groupsobserved by Boeri (2002) and Sapir (2006), plus one dummy for the NewMember States (NMS). Results are shown in Table 3, column 1, and alsoplotted in Figure 3. The geographical dummies are basically all significantand the goodness of fit is apparently high (adjusted R2 equal to 52 per cent).However a closer examination of Figure 3 shows that this result is reallydeceptive, as the observations for many countries in the different groupsoverlap with each other on the vertical scale.

In column 2 we introduce a set of partially ‘New Geographical Dum-mies’.13 They all have significant coefficients and the adjusted R2 increase to58 per cent. But also in this case there is a relevant overlap of the observationsfrom the different groups (see Figure 4).

From column 3 onwards we introduce additional regressors to measurelabour market institutions and policies. In particular we use two alternativespecifications of ALMP. In one case we simply scale policy expenditures byGDP (column 3); in the other we further divide it by the unemployment rate,for the reason discussed in Nickell (1997) (column 4). In both cases ALMPare non-significant. Instead, the rigidity index and the replacement rates arequite significant and correctly14 signed. Only one geographical dummy losessignificance, FTR (‘Fast Transition’), whereas the dummies for CTE (‘Con-tinental Enlarged’, which also includes some NMS), MEE ( ‘MediterraneanEnlarged’, which includes Cyprus and Malta) and PoRo (Poland andRomania, the two ‘misbehaving’ countries noted in Figure 2) retain theirsignificance. If we drop the geographical dummies entirely, the pattern ofsignificant institutions changes again (columns 5 and 6). Focusing on column6, where we use the ‘Nickell’ variable ([ALMP/GNP]/Unemployment), weobserve a strong positive coefficient for this variable. Also the short termreplacement rate and union density have significant positive effects, and thetax wedge a negative one.

However, even after dropping the geographical dummies, it is still neces-sary to account for country effects. We do this in the remaining columns 7 to10, where we use a Random Effect (RE) model. This is a least square modelcorrected for the fact that the two successive observations for each countrycannot be treated as independent random draws (see Nickell, 1997). More-over, as we also include some time-invariant regressors (see Appendix), wecannot use a Fixed Effect model. The RE model allows us to account for part

13 See Table A1 in the Appendix for a definition of these dummies (Nord, Ftr, PoRo, Cte, Mee).14 According to the institutionalist view.



Tabl

e3:

Dep

ende

ntV

aria

ble:

Em

ploy

men

tR

ate O

LS

Mod

elR

ando

mE

ffec

tM

odel

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

[AL

MP

/GD

P]

[AL

MP

/GD

P]

UN

EM

.[A

LM

P/G

DP

][A

LM

P/G

DP

]U

NE

M.

[AL

MP

/GD

P]

UN

EM

.[A

LM

P/G

DP

][A

LM

P/G

DP

]¥

RE

LIG

ION

[AL

MP

/GD

P]

¥E

TH

IC

Dum

_NM

S-7

.655

**(2

.817

)D

um_C

ont

-4.6

71*

(2.5

90)

Dum

_Med

-8.3

73**

(3.3

67)

Dum

_Nor

d3.

862

3.86

00.

125

1.91

2(2

.716

)(2

.759

)(3

.432

)(2

.744

)D

um_F

TR

-5.5

60**

-1.4

19-0

.766

(2.5

86)

(2.1

89)

(1.9

35)

Dum

_PoR

o-1

6.49

1***

-15.

836*

**-1

6.07

3***

(2.1

70)

(1.7

67)

(1.5

62)

Dum

_CT

E-5

.433

**-5

.993

*-5

.589

*(2

.519

)(2

.977

)(2

.868

)D

um_M

EL

-8.3

70**

-6.5

23**

-6.0

47**

(3.4

23)

(2.5

31)

(2.3

21)

Rig

idity

ofE

mpl

oym

ent

Inde

x-0

.159

***

-0.1

78**

*-0

.151

*-0

.125

-0.0

41-0

.067

-0.0

65-0

.044

(0.0

54)

(0.0

45)

(0.0

79)

(0.0

81)

(0.0

82)

(0.0

82)

(0.0

62)

(0.0

69)

Tax

Wed

ge-0

.083

-0.1

13-0

.362

**-0

.292

*-0

.238

-0.2

36-0

.265

-0.2

32(0

.157

)(0

.154

)(0

.150

)(0

.149

)(0

.203

)(0

.217

)(0

.193

)(0

.206

)U

nion

Den

sity

-0.0

15-0

.029

0.10

6**

0.09

6**

0.11

5**

0.13

6***

0.06

90.

072

(0.0

69)

(0.0

63)

(0.0

44)

(0.0

41)

(0.0

47)

(0.0

48)

(0.0

56)

(0.0

72)

Shor

tTe

rmR

epla

cem

ent

Rat

e0.

188*

**0.

186*

**0.

149*

*0.

123*

0.05

20.

105

0.06

60.

045

(0.0

66)

(0.0

66)

(0.0

67)

(0.0

64)

(0.0

93)

(0.0

98)

(0.0

82)

(0.0

95)

AL

MP

12.6

411.

748

3.83

725

.608

***

18.7

03**

*1.

098

5.58

5***

16.1

85**

*(7

.598

)(1

.253

)(2

.275

)(8

.807

)(5

.744

)(1

.322

)(1

.683

)(5

.947

)A

LM

P¥

Cul

ture

-8.8

63**

*-4

2.56

9***

(2.0

71)

(15.

631)

Cul

ture

-12.

750

(19.

560)

Cha

nge

inIn

flatio

n(%

p.a.

)0.

239

0.22

60.

494

0.47

90.

966*

0.80

8-0

.211

0.06

10.

066

0.19

3(0

.294

)(0

.378

)(0

.480

)(0

.501

)(0

.505

)(0

.486

)(0

.269

)(0

.252

)(0

.241

)(0

.282

)T

ime

Dum

my

YY

YY

YY

YY

YY

Obs

erva

tions

3939

3939

3939

3939

3939

Adj

uste

dR

-squ

ared

orR

2ov

eral

l0.

520.

580.

720.

700.

470.

550.

540.

460.

620.

59

RM

SE4.

123.

893.

193.

254.

334.

011.

261.

341.

211.

39N

umbe

rof

id23

2323

2323

2323

2323

23

Rob

ust

stan

dard

erro

rsin

pare

nthe

ses.

*si

gnifi

cant

at10

%;

**si

gnifi

cant

at5%

;**

*si

gnifi

cant

at1%

.Adj

uste

dR

-squ

ared

for

OL

Sm

odel

,R2

over

all

for

Ran

dom

Eff

ect

Mod

el.

Cul

ture

:=

Rel

igio

n(c

olum

n9)

,=W

ork-

Eth

ic(c

olum

n10

).So

urce

:Aut

hors

’ca

lcul

atio

ns.



of the unobserved (random) heterogeneity,15 and thus should give improvedcoefficient estimates. In column 7 we show the RE estimation of the sameequation as in column 6.16 The tax wedge and the replacement rates are nolonger significant, but the union density and ALMP retain their significanceand a similar magnitude. Similarly, in column 8 we show the RE estima-tion of the same equation as in column 5: again we find that ALMP is

15 If there is no correlation between observed country characteristics included in the regression (labourmarket institutions, change in inflation) and unobserved country effects (time-invariant), both fixed andrandom effect techniques provide consistent estimates of the coefficients in the equation. However, therandom effect estimation is more efficient, i.e. the standard errors of the estimators are lower. In fact thewithin-fixed effects ‘throw away’ variability, either by differencing (in the case of two observations acrosstime) or by computing the distance from the mean (in the case of more than two). The random effectestimation, on the other hand, simply accounts for the fact that the errors for paired observations in thepanel are correlated with one another. Hence, unpaired observations need to be weighted differently, inorder to account for heteroscedasticity. However this is only a theoretical possibility. We might suspect thatthere is correlation between observed country characteristics and unobserved country effects and, in thatcase, we should use FE.16 Fialova and Schneider (2008) also report results similar to those of column 7, in reference to a broadlyoverlapping sample of countries.

Figure 3: Traditional Geographical Dummies (Tab. 3, col. 1)

at at

bebe

cz czde de

dk dk

ee

es

es

fifi

fr

fr

gr

gr

hu

ie

ie

it

it

ltlu lulv

nlnl

pl

ptpt

sese

si

sk

uk uk

50

55

60

65

70

75

60 65 70 75Fitted values




non-significant, when it is only deflated by GDP. We suspect that this may berelated to the fact that ALMP have different effects, depending on a country’sinformal institutions and culture. Thus in columns 9 and 10 we additionallyinteract ALMP/GNP with a dummy variable, respectively ‘Religion’ (equal to1 if more than 50 per cent of the population has a Catholic or Orthodoxaffiliation) and ‘negative work ethic’ (defined as the percentage in a repre-sentative sample of the population in each country of those who agree withthe statement that ‘People should not have to work if they do not want to’).17

Column 9 shows that the relation between ALMP and employment rates isnow positive and significant for the non-Catholic and non-Orthodox countries(coefficient = +5.6) while for Catholic or Orthodox countries it is negative(the net effect is given by: 5.6 - 8.7 = -3.1). Column 10 shows a similarinteraction effect in countries where a high number of citizens support anegative work ethic. The interpretation of the regression coefficients in this

17 This variable is obtained from the World Value Survey, and takes higher values as the proportion of thosewho agree with this statement increases. See: «http://www.worldvaluessurvey.org/».

Figure 4: New Geographical Dummies (Tab. 3, col. 2)

atat

bebe

cz czdede

dkdk

ee

es

es

fifi

fr

fr

gr

gr

hu

ie

ie

it

it

ltlulu lv

nlnl

pl

ptpt

sese

si

sk

uk uk

50

55

60

65

70

75

50 55 60 65 70 75Fitted values




column is as follows: since the ratio of the absolute value of the two coeffi-cients (ALMP and ALPM x Culture) is equal to 38 per cent = 16.2/42.6, thisimplies that, in countries where the percentage of respondents who supporta ‘low’ work ethic is below 38 per cent,18 the relation between ALMP andemployment rates is positive. Vice versa, for countries where the percentageof respondents who support a low work ethic is above 38 per cent, the relationis negative. This result is in line with the idea that a social model based on amore intensive recourse to active policies would be well functioning in theScandinavian countries (but also in the UK, Slovenia and Slovakia, forexample). Figure 5 shows the fitted-actual graph for the RE model presentedin column 10. The lining up of actual values along the 45 degrees line can nowbe favourably compared with those in Figures 3 and 4.

As a notice of caution, however, we must also report that if we use eitherof the cultural interaction variables (religion; work ethic) in regressions where

18 These countries are: the Netherlands, the United Kingdom, Slovakia, Slovenia, Sweden, Denmark,Portugal, Hungary, Austria and Germany.

Figure 5: RE Model with Work Ethic (Tab. 3, col. 10)

at at

bebe

cz czde de

dkdk

ee

es

es

fifi

fr

fr

gr

gr

hu

ie

ie

it

it

ltlu lulv

nl nl

pl

ptpt

sese

si

sk

uk uk

50

55

60

65

70

75

55 60 65 70 75 80Fitted Values




ALMP is measured ‘à la Nickell’ (that is, deflated by unemployment rates),then we do not obtain significant results. We interpret this result as suggestingthat the ALMP/GDP/Unemployment ratio is a better and timelier measure ofthe counter-cyclical impact of labour market policies; whereas using thesimple ALMP/GDP ratio is a better indicator of a ‘structural’ policy stance,whose effectiveness is really conditioned by the overall framework of eachcountry’s formal and informal institutions.

Robustness Checks

In this section we present some additional regression results, as a robustnesscheck for the results in Table 3. First, we experiment with the use of alterna-tive measures of labour market expenditures. In Table 3 we used expenditureson ALMP according to the Eurostat classification. In Table 4 we reproduce(for ease of comparability) as columns 1, 4, 7 the results of columns 7, 9, 10of Table 3. In columns 2, 5, 8 instead we use expenditure on passive LMP(PLMP); in columns 3, 6, 9 total LMP (TLMP).19 In general there are nomajor differences in the sign and significance of the coefficients of LMP usingalternative measures, although (as it would be expected in the case of a‘causal’ effect running from policies to employment rates) the highest coef-ficient can always be associated with active policies.

Second, in Table 5 we modify the dependent variable, using participationrates instead of employment rates. We expect to get a weaker relation betweenLMP and the participation rate, as the latter includes the dynamics of bothunemployment and employment rates. LMP should enhance the overall par-ticipation rate; however they might impinge with a negative sign on unem-ployment and a positive sign on employment, and this is likely to reduce theestimated impact on the overall participation rate. Results line up with ourexpectations, and confirm those of Table 4, as the magnitude and statisticalsignificance of the coefficients in Table 5 are reduced, as expected, but retainthe same signs.

Third, in Table A2 of the Appendix we modify the sample in order toaccount for the largest available number of countries. Including 25 EU coun-tries in the regression allows us to exploit 41 observations. The statisticalsignificance of all the coefficients on LMP is even stronger. We also confirmthe evidence that LMP and employment levels are non-monotonically related,by accounting for cross-country differences in ‘culture’ (religion and workethic). However in these regressions we are unable to include measures forunion density and for the short term replacement rate, as these data areunavailable for the added countries.

19 See the Appendix for the relation between policy types and the Eurostat classification.



Tabl

e4:

Dep

ende

ntV

aria

ble:

Em

ploy

men

tR

ate.

Alte

rnat

ive

Mea

sure

sfo

rL

abou

rM

arke

tE

xpen

ditu

res

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

[LM

P/G

DP

]U

NE

M.

[LM

P/G

DP

]¥

Rel

igio

n[L

MP

/GD

P]

¥W

ork-

Eth

icA

LM

PP

LM

PT

LM

PA

LM

PP

LM

PT

LM

PA

LM

PP

LM

PT

LM

P

Rig

idity

ofE

mpl

oym

ent

Inde

x-0

.041

-0.0

41-0

.027

-0.0

65-0

.068

-0.0

63-0

.044

-0.0

39-0

.037

(0.0

82)

(0.0

85)

(0.0

87)

(0.0

62)

(0.0

71)

(0.0

69)

(0.0

69)

(0.0

78)

(0.0

80)

Tax

Wed

ge-0

.238

-0.1

31-0

.144

-0.2

65-0

.176

-0.2

11-0

.232

-0.1

98-0

.206

(0.2

03)

(0.2

25)

(0.2

17)

(0.1

93)

(0.2

14)

(0.2

02)

(0.2

06)

(0.2

08)

(0.2

09)

Uni

onD

ensi

ty0.

115*

*0.

105*

*0.

094*

*0.

069

0.08

8*0.

078

0.07

20.

078

0.07

1(0

.047

)(0

.047

)(0

.046

)(0

.056

)(0

.054

)(0

.053

)(0

.072

)(0

.055

)(0

.060

)Sh

ort

Term

Rep

lace

men

tR

ate

0.05

20.

027

0.00

40.

066

0.11

40.

094

0.04

50.

034

0.03

7(0

.093

)(0

.103

)(0

.101

)(0

.082

)(0

.102

)(0

.096

)(0

.095

)(0

.098

)(0

.099

)L

MP

18.7

03**

*15

.757

**11

.654

***

5.58

5***

2.04

11.

693*

*16

.185

***

7.67

36.

085*

(5.7

44)

(6.4

38)

(4.1

74)

(1.6

83)

(1.2

84)

(0.7

64)

(5.9

47)

(6.1

60)

(3.6

85)

LM

P¥

Cul

ture

-8.8

63**

*-3

.916

***

-2.7

25**

*-4

2.56

9***

-17.

085

-14.

360

(2.0

71)

(1.4

97)

(0.8

16)

(15.

631)

(15.

742)

(9.3

64)

Cul

ture

-12.

750

-17.

349

-10.

840

(19.

560)

(23.

163)

(23.

307)

Cha

nge

inIn

flatio

n(%

p.a.

)-0

.211

-0.1

63-0

.296

0.06

60.

022

0.03

20.

193

0.20

10.

145

(0.2

69)

(0.2

26)

(0.2

36)

(0.2

41)

(0.2

18)

(0.2

21)

(0.2

82)

(0.2

74)

(0.2

50)

Tim

eD

umm

yY

YY

YY

YY

YY

Obs

erva

tions

3939

3939

3939

3939

39N

umbe

rof

id23

2323

2323

2323

2323

R2

over

all

0.54

0.54

0.55

0.62

0.57

0.60

0.59

0.62

0.61

RM

SE1.

261.

121.

111.

211.

271.

311.

391.

451.

43

Rob

ust

stan

dard

erro

rsin

pare

nthe

ses.

*si

gnifi

cant

at10

%;

**si

gnifi

cant

at5%

;**

*si

gnifi

cant

at1%

.C

ultu

re:

=R

elig

ion

(col

umns

4–5–

6),

=W

ork-

Eth

ic(c

olum

ns7–

8–9)

.So

urce

:Aut

hors

’ca

lcul

atio

ns.



Tabl

e5:

Dep

ende

ntV

aria

ble:

Part

icip

atio

nR

ate.

Alte

rnat

ive

Mea

sure

sfo

rL

abou

rM

arke

tE

xpen

ditu

res

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

[LM

P/G

DP

]U

NE

M.

[LM

P/G

DP

]¥

Rel

igio

n[L

MP

/GD

P]

¥W

ork-

Eth

icA

LM

PP

LM

PT

LM

PA

LM

PP

LM

PT

LM

PA

LM

PP

LM

PT

LM

P

Rig

idity

ofE

mpl

oym

ent

Inde

x-0

.058

-0.0

68-0

.059

-0.0

64-0

.066

-0.0

53-0

.055

-0.0

47-0

.042

(0.0

71)

(0.0

68)

(0.0

71)

(0.0

55)

(0.0

63)

(0.0

59)

(0.0

63)

(0.0

71)

(0.0

73)

Tax

Wed

ge-0

.076

-0.0

04-0

.012

-0.1

470.

012

-0.0

85-0

.095

-0.0

25-0

.066

(0.1

85)

(0.2

01)

(0.1

94)

(0.1

67)

(0.2

01)

(0.1

75)

(0.2

12)

(0.2

18)

(0.2

11)

Uni

onD

ensi

ty0.

091*

*0.

087*

*0.

080*

*0.

055

0.04

60.

044

0.06

70.

065

0.06

0(0

.041

)(0

.041

)(0

.041

)(0

.046

)(0

.047

)(0

.045

)(0

.065

)(0

.052

)(0

.054

)Sh

ort

Term

Rep

lace

men

tR

ate

0.04

70.

042

0.02

80.

047

0.06

20.

048

0.04

10.

033

0.03

9(0

.094

)(0

.097

)(0

.099

)(0

.073

)(0

.092

)(0

.077

)(0

.101

)(0

.115

)(0

.103

)L

MP

10.7

19**

7.64

3*5.

976*

4.66

1***

2.61

2***

1.99

2***

13.7

32**

*6.

643

6.14

7*(5

.192

)(4

.291

)(3

.079

)(1

.584

)(1

.012

)(0

.556

)(4

.589

)(5

.906

)(3

.352

)L

MP

¥C

ultu

re-6

.406

***

-3.6

77**

*-2

.400

***

-34.

774*

**-1

3.75

9-1

3.94

9(1

.659

)(1

.158

)(0

.614

)(1

1.65

2)(1

5.67

2)(8

.740

)C

ultu

re3.

004

-1.0

838.

122

(16.

150)

(22.

903)

(20.

778)

Cha

nge

inIn

flatio

n(%

p.a.

)-0

.329

-0.2

63-0

.341

-0.1

44-0

.259

-0.2

20-0

.196

-0.2

04-0

.235

(0.2

20)

(0.1

93)

(0.2

10)

(0.2

36)

(0.1

65)

(0.1

81)

(0.1

98)

(0.1

66)

(0.1

58)

Tim

eD

umm

yY

YY

YY

YY

YY

Obs

erva

tions

3939

3939

3939

3939

39N

umbe

rof

id23

2323

2323

2323

2323

R2

over

all

0.45

0.43

0.45

0.65

0.57

0.63

0.50

0.49

0.51

RM

SE0.

920.

890.

880.

950.

860.

950.

930.

960.

98

Rob

ust

stan

dard

erro

rsin

pare

nthe

ses.

*si

gnifi

cant

at10

%;

**si

gnifi

cant

at5%

;**

*si

gnifi

cant

at1%

.C

ultu

re:

=R

elig

ion

(col

umns

4–5–

6),

=W

ork-

Eth

ic(c

olum

ns7–

8–9)

.So

urce

:Aut

hors

’ca

lcul

atio

ns.



Finally, in Table A3 of the Appendix we introduce a measure of thestrictness of ‘Employment Protection Laws’ (EPL, source: OECD) in place ofthe ‘Rigidity of Employment Index’ (World Bank). In addition, we alsoinclude an indicator of the strictness of ‘Product Market Regulation’ (source:OECD). No major differences are detected on the impact of LMP on thedependent variable. The index of product market regulation is negativelyassociated with the employment level and EPL is insignificant. However inthis case the number of countries is only 18 and the sample size drops to 31observations, impinging both upon the efficiency of the estimators and theavailable degrees of freedom.

Conclusions

In this article we have explored the relations between employment rates andseveral indicators of labour market institutions and policies. Our studyfocused on the EU and is based on macro variables that cover between 23 and25 countries. Using the same or a comparable set of indicators as in Nickell(1997) and Blanchard and Wolfers (2000), we have observed that expendi-tures on labour market programmes, and especially on active programmes,are significantly positively related to employment levels. Other variablesoften used in previous research are found to have much less significant andstable effects, except possibly for union density, which in some specificationsseems also to be positively correlated to employment levels; on the other handthis variable is significantly higher in the three Scandinavian countries, andthus may be really proxying for the specific features of their labour marketinstitutions.

We have also confirmed that the relationship between ALMP andemployment rates is non-monotonic. We interpret this to imply that ALMPmay have a positive impact on employment especially in those countrieswhere informal institutions support a strong ‘pro-work’, and possibly amore individualistic attitude; vice versa, in countries where these attitudesare not widely shared there is no indication that ALMP are positivelyrelated to employment rates. Similarly, we have found that the relationbetween ALMP and employment rates is negative in countries with a pre-vailing Catholic or Orthodox affiliation, while it is significantly positive inthe other countries. We suggest that these results may be interpreted in thelight of the previous findings of Bassanini and Duval (2006), that ‘structuralreforms appear to have mutually reinforcing effects: the impact of a givenpolicy reform is greater the more employment-friendly the overall policyand institutional framework’. Our results suggest that the notion of an



institutional framework should be enlarged, to include the role of informalinstitutions.

As a note of caution, we acknowledge that the aggregate nature of ouranalysis limits the confidence in the robustness and in the interpretation of ourresults. However, and for the same reason, these results contribute to weakenthe conclusions of previous studies sharing the same methodology, on therole of specific labour market polices and institutions, such as measures ofemployment rigidity, tax wedge and union density. We suggest that thesevariables may really be acting as confounding factors for other informalinstitutions or country characteristics.

Also, we have tentatively suggested that analyses based on the ‘macro’framework adopted by Nickell (1997) and others, and those which follow amore ‘institutionalist’ approach, focused on the variety of labour marketmodels, can be meaningfully integrated in the same empirical framework.When this is done, informal institutions more than simple ‘geography’ appearto embody the relevant discriminant factors to account for the observeddifferences in the performance of labour markets.

Correspondence:Riccardo RovelliDipartimento di Scienze EconomicheStrada Maggiore 4540125 Bologna, ItalyTel +39.051.209.2601Fax +39.051.209.2664email [email protected]

References

Bassanini, A. and Duval, R. (2006) ‘Employment Patterns in OECD Countries:Reassessing the Role of Policies and Institutions’. OECD Economics DepartmentWorking Paper No. 486.

Betcherman, G., Olivas, K. and Dar, A. (2004) ‘Impacts of Active Labour MarketProgrammes: New Evidence from Evaluations with Particular Attention to Devel-oping and Transition Countries’. World Bank – Social Protection DiscussionPaper No. 0402.

Blanchard, O. and Wolfers, J. (2000) ‘The Role of Shocks and Institutions in the Riseof European Unemployment: The Aggregate Evidence’. The Economic Journal,Vol. 110, C1–C33.

Boeri, T. (2002) ‘Let Social Policy Models Compete and Europe Will Win’. Paperpresented at a Conference hosted by the Kennedy School of Government, HarvardUniversity, 11–12 April 2002.



Bradley, D.H. and Stephens, J.D. (2007) ‘Employment Performance in OECD Coun-tries. A test of Neoliberal and Institutionalist Hypotheses’. Comparative PoliticalStudies, Vol. 40, No. 12, pp. 1486–510.

Checchi, D. and Lucifora, C. (2002) ‘Unions and Labour Market Institutions inEurope’. Economic Policy, Vol. 17, No. 35, pp. 363–408.

Conway, P., Janod, V. and Nicoletti, G. (2005) ‘Product Market Regulation in OECDCountries, 1998 to 2003’. OECD Economics Department Working Paper, No.419.

Dar, A. and Tzannatos, Z. (1999) ‘Active Labour Market Programmes: A Review ofthe Evidence from Evaluations’. World Bank – Social Protection DiscussionPaper No. 99.01.

Esping-Andersen, G. (1990) The Three Worlds of Welfare Capitalism (Princeton, NJ:Princeton University Press).

Eurobarometer (2007) ‘European Social Reality’. Special Eurobarometer 273,February. Available at «http://ec.europa.eu/public_opinion/archives/ebs/ebs_273_en.pdf».

Ferrera, M. (1998) ‘The Four “Social Europes”: Between Universalism and Selec-tivity’. In Rhodes, M. and Meny, Y. (eds) The Future of European Welfare: A NewSocial Contract? (Basingstoke: Macmillan).

Fialova, K. and Schneider, O. (2008) ‘Labour Market Institutions and their Effect onLabour Market Performance in the New EU Member Countries’. Cesifo WorkingPaper, No. 2421, October.

Heckman, J. and Pagés, C. (2003) ‘Law and Employment: Lessons from LatinAmerica and the Caribbean’. NBER Working Paper No. 10129.

Hicks, A. and Kenworthy, L. (2003) ‘Varieties of Welfare Capitalism’. Socio-Economic Review, Vol. 1, pp. 27–61.

Howell, D.R., Baker, D., Glyn, A. and Schmitt, J. (2007) ‘Are ProtectiveLabour Market Institutions at the Root of Unemployment? A Critical Review ofthe Evidence’. Capitalism and Society, Vol. 2, No. 1.

Nicoletti, G. and Scarpetta, S. (2005) ‘Regulation and Economic Performance:Product Market Reforms and Productivity in the OECD’. OECD EconomicsDepartment Working Paper No. 460.

Nickell, S. (1997), ‘Unemployment and Labour Market Rigidities: Europeversus North America’. The Journal of Economic Perspectives, Vol. 11, No. 3, pp.55–74.

Nickell, S., Nunziata, L. and Ochel, W. (2005), ‘Unemployment in the OECDsince the 1960s: What Do We Know?’ The Economic Journal, Vol. 115,pp. 1–27.

Rovelli, R. (2008) ‘One Market, How many Social Models? Policy and PerformanceIndicators in the EU’. In Luboš, T., Plechanovová, B. and Schneider, O. (eds)European Integration at the Crossroads (Prague: Universitas Carolina FacultasIuridica).

Rovelli, R. and Bruno, R.L. (2008) ‘Labour Market Policies, Institutions andEmpolyment Rates in the EU-27’. IZA Discussion Paper, No. 3502.



Sapir, A. (2006) ‘Globalization and the Reform of European Social Models’. JCMS,Vol. 44, No. 2, pp. 369–90.

Visser, J. (2006) ‘Union Membership Statistics in 24 Countries’. Monthly LabourReview, Vol. 129, No. 1, pp. 38–49.

Supporting Information

Additional Supporting Information may be found in the online version of thisarticle:

Appendix. The Data.

Please note: Wiley-Blackwell are not responsible for the content or function-ality of any supporting materials supplied by the authors. Any queries (otherthan missing material) should be directed to the corresponding author for thearticle.



labor market policies, institutions and employment rates in the eu-27

Documents